Alexey Solovey
depends on what you are going to classify. Some actions of cartoon vs character movie?
https://arxiv.org/pdf/1604.06573.pdf
and
https://arxiv.org/pdf/1412.0767.pdf
and:
you can create a convolutional 2D->Flatten high-level features set and put these features as an input for RNN.
So the artitecture can be mixed:
For each frame: i.e:
1. (Conv2D(3,3)->Con2D(3,3)->MaxPool)(N)->Flatten-> {vector of high-level featires}
2. Put this vector to LSTMs / RNN
Conn2D architecture can be from pre-trainet models like VGG19 / Inception
0 votes
0 thanks
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification
More Hany El-Ghaish's questions See All
Similar questions and discussions